generated: '2026-07-21'
method: searched
source: https://docs.topk.io/changelog
scheme: monthly-dated
current_sdk_version: '0.12.0'
docs: https://docs.topk.io/changelog
entries:
- date: '2026-06'
  breaking: false
  highlights:
  - Added Partitioned Collections (Namespaces) for effortless multi-tenancy with fully isolated, on-demand partitions.
  - Made SQL the search query language for hybrid and multi-vector search (no proprietary query language required).
  - Introduced semantic_index enabling production-ready multi-vector retrieval with a single schema annotation.
  - Optimized the SMVE kernel to match GPU performance with just 8 CPU threads.
- date: '2026-05'
  breaking: false
  highlights:
  - Added native support for structs with full indexing and filtering (search/filter nested structured data without flattening the schema).
  - Released Iso-ModernColBERT, a late-interaction model optimized for efficient inference and scalable retrieval.
- date: '2026-04'
  breaking: false
  highlights:
  - Shipped the Datasets API for unstructured document ingestion, search, and grounded question answering (PDF, Markdown, HTML, and more).
  - Shipped the CLI to manage datasets, ingest documents, and query them from the terminal.
  - Shipped the MCP server to connect private data inside datasets to AI agents.
- date: '2026-03'
  breaking: false
  highlights:
  - Added support for tuning BM25 (configure b and k1 parameters to adjust ranking behavior).
  - Published a blog post about SMVE, an algorithm for sparse/multi-vector execution.
older_entries_window: Changelog documents monthly entries back to June 2025; see the docs for the full history.